Re: help to (slightly) modify Lua interpreter

[Leo Razoumov <slonik.az@...>]

On 2009-11-06, Roberto Ierusalimschy <roberto@inf.puc-rio.br> wrote:
> > I'm planning to translate the code like that:
>  >
>  > m[i, j]         => __index(m, i, j)
>  > m[i, j] = v    => __newindex(m, i, j, v)
>  >
>  > So all the indexes will be feeded to the __index or __newindex
>  > metamethods in the order. I guess that the variable number of
>  > arguments is not a problem: the metamethods are supposed to check how
>  > many arguments and act accordingly.
>
>
> I guess the next code gives a reasonable solution for 2D arrays.
>  (It should not be too dificult to generalize it to n dimensions.)
>
>  -------------------------------------
>  local mt1 = { __index = function (t,k)
>                 local nt = setmetatable({}, {
>                              __index = function (t1, k1)
>                                          return t:index(k,k1)
>                                        end,
>                              __newindex = function (t1, k1, v)
>                                             t:newindex(k,k1,v)
>                                           end,
>                            })
>                 t[k] = nt
>                 return nt
>               end,
>               mode = 'v',
>             }
>
>
>  function newmatrix (index, newindex)
>   return setmetatable({index = index, newindex = newindex}, mt1)
>  end
>  -------------------------------------

Every time m[i][j] is called two new closures are created to be used
only once. Would it drive GC crazy if m[i][j] is used withing loops
millions of times??

--Leo--